\chapter{Coherent states and a classical block dynamics}\label{s:quantum} In this section, we compare the spectral gap of $H_{L, N}$ with the spectral gap of a classical Markov dynamics, the block dynamics of a probability measure $\mu_{L, N}$ on $(\Sph)^\Lambda$. The comparison is \zcref{t:comparison}. In the proof of \zcref{t:main}, it is combined with the lower bound on the gap of the block dynamics proved in \zcref{s:block} and with the explicit lower bound on the constant $g_b$ of \zcref{t:comparison} proved in \zcref{l:patchgapexplicit}. We also identify the kernel of $H_{L, N}$ in \zcref{c:kernel}. The heuristics are explained in \zcref{s:ov-quantum}. The polynomial representation of spin $2$, the VBS state, and the classical measure that the VBS state defines through spin coherent states go back to \cite{AKLT88, AAH88, KLT88}; we recall them in \zcref{s:quantum-poly,s:quantum-coherent} in the form that we use. In \zcref{s:quantum-rotated}, we introduce the measures on finite subsets of $\Z^2$ that are used in the rest of the paper, and we relate them to the conditional laws of $\mu_{L, N}$. Throughout this section, $L, N \ge 3$ are integers, and $\Lambda = \Lambda_{L, N}$ and $\mathcal E$ are as in \zcref{s:intro-result}. Since $L, N \ge 3$, every site of $\Lambda$ belongs to exactly four bonds of $\mathcal E$. For $x \in \Lambda$ and $u \in \Z^2$, we write $x + u$ for the sum in $\Lambda$, in which $u$ is reduced modulo $(L, N)$. For $R \subset \Lambda$, we let $\mathcal E(R)$ denote the set of bonds of $\mathcal E$ that meet $R$. Given a configuration $s = (s_x)_{x \in \Lambda}$ and a set $D \subset \Lambda$, we let $s_D = (s_x)_{x \in D}$ denote the restriction of $s$ to $D$, and we use the same notation for configurations on subsets of $\Z^2$. \section{The polynomial representation}\label{s:quantum-poly} We begin with spin-$\frac12$ coherent states. Let $\sigma = (\sigma_1, \sigma_2, \sigma_3)$ denote the vector of Pauli matrices. For $a = (a_1, a_2, a_3) \in \Sph$, we set \begin{equation*} \zeta(a) = \frac{1}{\sqrt{2 (1 + a_3)}} \bigl( 1 + a_3, a_1 + \iu a_2 \bigr) \in \C^2 \quad \text{if } a_3 \neq -1, \qquad \zeta \bigl( (0, 0, -1) \bigr) = (0, 1) . \end{equation*} Then $\zeta \colon \Sph \to \C^2$ is measurable, $\zeta(a)$ is a unit vector, and a direct computation shows that \begin{equation}\label{e:zeta} \zeta(a) \zeta(a)^* = \frac12 (\Id + a \cdot \sigma) \qquad \text{for all } a \in \Sph . \end{equation} Every orthogonal projection of rank one on $\C^2$ has the form $\frac12 (\Id + a \cdot \sigma)$ for exactly one $a \in \Sph$. Since such a projection determines the unit vectors in its range up to a factor of modulus one, every unit vector of $\C^2$ has the form $e^{\iu \alpha} \zeta(a)$ with $a \in \Sph$ and $\alpha \in \R$. For $g \in \mathrm{SU}(2)$, we let $\bar g \in \mathrm{SO}(3)$ denote the rotation defined by $g (a \cdot \sigma) g^* = (\bar g a) \cdot \sigma$ for all $a \in \R^3$. By \eqref{e:zeta}, the unit vectors $g \zeta(a)$ and $\zeta(\bar g a)$ have the same projection $\frac12 (\Id + (\bar g a) \cdot \sigma)$, and $g \zeta(a) = e^{\iu \alpha} \zeta(\bar g a)$ for some $\alpha \in \R$. We now realize the spin-$2$ representation. Let $V$ denote the space of homogeneous polynomials of degree $4$ in $z = (z_1, z_2) \in \C^2$, with the inner product \begin{equation}\label{e:Vinner} \langle \varphi, \psi \rangle_V = 5 \int_{\Sph} \overline{\varphi(\zeta(a))} \, \psi(\zeta(a)) \dd a . \end{equation} Since $\varphi$ and $\psi$ are homogeneous of the same degree, the integrand is a polynomial in the entries of $\zeta(a) \zeta(a)^*$. In particular, it is continuous in $a$ and unchanged when $\zeta(a)$ is multiplied by a complex number of modulus one. The group $\mathrm{SU}(2)$ acts on $V$ by $(U_g \varphi)(z) = \varphi(g^{-1} z)$. Since $g^{-1} \zeta(a)$ is a multiple of $\zeta(\bar g^{-1} a)$ by a complex number of modulus one and $\dd a$ is invariant under rotations, the inner product \eqref{e:Vinner} is invariant under this action. To see that it is positive definite, suppose that $\langle \varphi, \varphi \rangle_V = 0$. Then $\varphi(\zeta(a)) = 0$ for all $a$ by continuity, and $\varphi$ vanishes on the unit sphere of $\C^2$, since every unit vector has the form $e^{\iu \alpha} \zeta(a)$. By homogeneity, $\varphi = 0$. Recall that the action of $\mathrm{SU}(2)$ on the homogeneous polynomials of degree $2j$ in two variables is the irreducible representation of spin $j$. We use the realization $V$ of spin $2$ throughout, and we identify $\mathcal H_\Lambda = \bigotimes_{x \in \Lambda} V_x$ with the space of polynomials in $z_\Lambda = (z_x)_{x \in \Lambda}$, where $z_x = (z_{x, 1}, z_{x, 2}) \in \C^2$, that are homogeneous of degree $4$ in each $z_x$. Here the product vector $\bigotimes_x \varphi_x$ corresponds to the polynomial $\prod_x \varphi_x(z_x)$, and the inner product of $\mathcal H_\Lambda$ is the product of the inner products \eqref{e:Vinner}. Recall from \zcref{s:intro-result} that, for a bond $e = \{x, y\}$, $P_e$ is the orthogonal projection of $V_x \otimes V_y$ onto its subspace of total spin $4$, extended by the identity. In our realization, $\mathrm{SU}(2)$ acts on $V_x \otimes V_y$ by $U_g \otimes U_g$. By the Clebsch--Gordan rule, $V_x \otimes V_y$ is the orthogonal direct sum of irreducible subspaces of spins $0, 1, 2, 3, 4$, each of which occurs once, and $P_e$ projects onto the subspace of spin $4$. For $R \subset \Lambda$, we set \begin{equation}\label{e:HR} H_R = \sum_{e \in \mathcal E(R)} P_e , \end{equation} and we let $\Pi_R$ denote the orthogonal projection of $\mathcal H_\Lambda$ onto $\ker H_R$. Note that $H_\Lambda = H_{L, N}$. We fix an orientation $(x, y)$ of every bond $e = \{x, y\}$, and we define the \emph{singlet polynomial} of $e$ by \begin{equation*} q_e = z_{x, 1} z_{y, 2} - z_{x, 2} z_{y, 1} . \end{equation*} (The other orientation gives $-q_e$.) Since $\det g = 1$ for $g \in \mathrm{SU}(2)$, we have \begin{equation}\label{e:qinv} q_e(g z_x, g z_y) = q_e(z_x, z_y) \qquad \text{for all } g \in \mathrm{SU}(2) . \end{equation} For a set $\mathcal F$ of bonds, we set $q_{\mathcal F} = \prod_{e \in \mathcal F} q_e$, and we call \begin{equation}\label{e:VBS} \VBS = q_{\mathcal E} = \prod_{e \in \mathcal E} q_e \end{equation} the \emph{VBS state}. Since every site belongs to four bonds, $\VBS$ is homogeneous of degree $4$ in each $z_x$, so that $\VBS \in \mathcal H_\Lambda$. Further, $\VBS \neq 0$, since the polynomial ring $\C[z_\Lambda]$ is an integral domain. \section{The kernel of the Hamiltonian}\label{s:quantum-kernel} The description of the kernel of $H_{L, N}$ in \zcref{c:kernel} is standard (see \cite{KLT88, KirillovKorepin89}). We include the short proofs of the two lemmas that give it, because we will also use \zcref{l:divisibility} with $R$ a block in the proof of \zcref{t:comparison}. \begin{lemma}\label{l:edge} Let $e = \{x, y\} \in \mathcal E$ and $\varphi \in \mathcal H_\Lambda$. Then $P_e \varphi = 0$ if and only if $\varphi = q_e \chi$ for a polynomial $\chi$ in $z_\Lambda$ that is homogeneous of degree $3$ in $z_x$ and in $z_y$ and of degree $4$ in $z_u$ for every $u \in \Lambda \setminus \{x, y\}$. \end{lemma} \begin{proof} Let $W_4$ and $W_3$ denote the spaces of polynomials in $(z_x, z_y)$ that are homogeneous of degree $4$, respectively $3$, in each of $z_x$ and $z_y$, so that $W_4 = V_x \otimes V_y$. The group $\mathrm{SU}(2)$ acts on $W_4$ and on $W_3$ by $\chi \mapsto \chi(g^{-1} z_x, g^{-1} z_y)$. We have \begin{equation*} \mathcal H_\Lambda = W_4 \otimes \mathcal H' , \end{equation*} where $\mathcal H'$ is the space of polynomials in $(z_u)_{u \neq x, y}$ that are homogeneous of degree $4$ in each $z_u$, and $P_e$ acts on the first factor. The polynomials $\chi$ in the statement are the elements of $W_3 \otimes \mathcal H'$. It therefore suffices to show that \begin{equation*} \{ \chi \in W_4 : P_e \chi = 0 \} = q_e W_3 . \end{equation*} Multiplication by $q_e$ maps $W_3$ into $W_4$ injectively, since $\C[z_x, z_y]$ is an integral domain. The invariance \eqref{e:qinv} of $q_e$ shows that this map commutes with the action of $\mathrm{SU}(2)$. The space of homogeneous polynomials of degree $3$ in $z \in \C^2$ is the irreducible representation of spin $\frac32$. By the Clebsch--Gordan rule, $W_3$ is the direct sum of irreducible subspaces of spins $0, 1, 2, 3$, and $W_4$ is the direct sum of irreducible subspaces of spins $0, \dots, 4$, each occurring once. We conclude that $q_e W_3$ is an invariant subspace of $W_4$ that is a sum of irreducible subspaces of spins $0, 1, 2, 3$. Since each spin occurs only once in $W_4$, the subspace $q_e W_3$ coincides with the sum of the components of $W_4$ of spins $0, 1, 2, 3$. This sum is the orthogonal complement of the component of spin $4$, which is the kernel of $P_e$ in $W_4$. \end{proof} \begin{lemma}\label{l:divisibility} Let $R \subset \Lambda$, and let $\varphi \in \mathcal H_\Lambda$ satisfy $P_e \varphi = 0$ for all $e \in \mathcal E(R)$. Then $\varphi = q_{\mathcal E(R)} \chi$ for a polynomial $\chi$ in the variables $(z_u)_{u \in \Lambda \setminus R}$. \end{lemma} \begin{proof} We may assume that $\varphi \neq 0$. We first show that each $q_e$ is irreducible in $\C[z_\Lambda]$. A nontrivial factorization of the homogeneous quadratic polynomial $q_e$ would write it as a product of two linear forms. Such a product is a quadratic form of rank at most $2$, whereas $q_e$ has rank $4$ as a quadratic form in $z_{x, 1}, z_{x, 2}, z_{y, 1}, z_{y, 2}$. If $e \neq e'$, then $q_e$ and $q_{e'}$ involve different sets of variables and are not scalar multiples of each other. By \zcref{l:edge}, $q_e$ divides $\varphi$ for every $e \in \mathcal E(R)$. In the unique factorization domain $\C[z_\Lambda]$, it follows that the product $q_{\mathcal E(R)}$ of these pairwise non-associated irreducible polynomials divides $\varphi$. Let $\chi$ denote the quotient. It remains to show that $\chi$ has degree $0$ in $z_x$ for all $x \in R$. Let $x \in R$, and let $\deg_{z_x}$ denote the degree in the pair of variables $z_x = (z_{x, 1}, z_{x, 2})$, which is additive under multiplication in $\C[z_\Lambda]$. The four bonds that contain $x$ belong to $\mathcal E(R)$, and each of them contributes a factor of degree $1$ in $z_x$ to $q_{\mathcal E(R)}$, while the other factors have degree $0$ in $z_x$. Since $\varphi$ has degree $4$ in $z_x$, we obtain \begin{equation*} 4 = \deg_{z_x} \varphi = \deg_{z_x} q_{\mathcal E(R)} + \deg_{z_x} \chi = 4 + \deg_{z_x} \chi , \end{equation*} which gives $\deg_{z_x} \chi = 0$. \end{proof} \begin{corollary}\label{c:kernel} For all integers $L, N \ge 3$, we have $\ker H_{L, N} = \C \VBS$. \end{corollary} \begin{proof} Let $e = \{x, y\} \in \mathcal E$. We have $\VBS = q_e \, q_{\mathcal E \setminus \{e\}}$, where $q_{\mathcal E \setminus \{e\}}$ is homogeneous of degree $3$ in $z_x$ and in $z_y$ and of degree $4$ in every other $z_u$. Then \zcref{l:edge} implies that $P_e \VBS = 0$. As $e$ is arbitrary, $H_{L, N} \VBS = 0$. Conversely, let $\varphi \in \ker H_{L, N}$. Then \begin{equation*} \sum_{e \in \mathcal E} \| P_e \varphi \|^2 = \langle \varphi, H_{L, N} \varphi \rangle = 0 , \end{equation*} so that $P_e \varphi = 0$ for all $e \in \mathcal E$. Applying \zcref{l:divisibility} with $R = \Lambda$, for which $\mathcal E(R) = \mathcal E$, we find that $\varphi = q_{\mathcal E} \chi = \chi \VBS$ for a constant $\chi$. \end{proof} \section{The coherent-state measure}\label{s:quantum-coherent} For a polynomial $p$ in $z_\Lambda$ and a configuration $s \in (\Sph)^\Lambda$, we write $p(\zeta(s)) = p \bigl( (\zeta(s_x))_{x \in \Lambda} \bigr)$. We define \begin{equation*} \iota \colon \mathcal H_\Lambda \to L^2 \Bigl( (\Sph)^\Lambda, \prod_{x \in \Lambda} 5 \dd s_x \Bigr), \qquad (\iota \varphi)(s) = \varphi(\zeta(s)) . \end{equation*} By \eqref{e:Vinner} and the definition of the inner product of $\mathcal H_\Lambda$, the identity $\langle \iota \varphi, \iota \psi \rangle = \langle \varphi, \psi \rangle$ holds for product vectors $\varphi$ and $\psi$. Since both sides are sesquilinear and the product vectors span $\mathcal H_\Lambda$, it holds for all $\varphi, \psi \in \mathcal H_\Lambda$, and $\iota$ is an isometry. We next compute $|\iota \VBS|^2$. Let $a, a' \in \Sph$. The number $\zeta(a)_1 \zeta(a')_2 - \zeta(a)_2 \zeta(a')_1$ is the determinant of the $2 \times 2$ matrix with columns $\zeta(a)$ and $\zeta(a')$. Its squared modulus is the Gram determinant $1 - |\langle \zeta(a'), \zeta(a) \rangle|^2$ of these unit vectors. By \eqref{e:zeta} and the identities $\Tr \sigma_j = 0$ and $\Tr (\sigma_j \sigma_k) = 2 \delta_{jk}$, we have \begin{equation*} |\langle \zeta(a'), \zeta(a) \rangle|^2 = \Tr \bigl( \zeta(a) \zeta(a)^* \zeta(a') \zeta(a')^* \bigr) = \frac14 \Tr \bigl( (\Id + a \cdot \sigma)(\Id + a' \cdot \sigma) \bigr) = \frac{1 + a \cdot a'}{2} . \end{equation*} Taking $a = s_x$ and $a' = s_y$, we find that every bond $e = \{x, y\}$ satisfies \begin{equation}\label{e:qmod} |q_e(\zeta(s))|^2 = \frac{1 - s_x \cdot s_y}{2} . \end{equation} We define the probability measure $\mu_{L, N}$ on $(\Sph)^\Lambda$ by \begin{equation}\label{e:mu} \mu_{L, N}(\dd s) = \frac{1}{Z_{L, N}} \prod_{\{x, y\} \in \mathcal E} (1 - s_x \cdot s_y) \prod_{x \in \Lambda} \dd s_x , \qquad Z_{L, N} = \int \prod_{\{x, y\} \in \mathcal E} (1 - s_x \cdot s_y) \prod_{x \in \Lambda} \dd s_x . \end{equation} By \eqref{e:VBS} and \eqref{e:qmod}, we have \begin{equation*} \mu_{L, N}(\dd s) = \frac{|\iota \VBS(s)|^2}{\| \VBS \|^2} \prod_{x \in \Lambda} 5 \dd s_x , \qquad Z_{L, N} = 2^{|\mathcal E|} \, 5^{-|\Lambda|} \, \| \VBS \|^2 > 0 . \end{equation*} The density in \eqref{e:mu} is positive outside the set of configurations with $s_x = s_y$ for some bond $\{x, y\}$, which has measure zero. In particular, $\mu_{L, N}$ and the product measure $\prod_x \dd s_x$ have the same null sets, and $\iota \VBS \neq 0$ almost everywhere. We let $L^2(\mu_{L, N})$ denote the space of complex-valued functions that are square integrable with respect to $\mu_{L, N}$. \begin{lemma}\label{l:coherent} Let $\varphi \in \mathcal H_\Lambda$, and let $F = \iota \varphi / \iota \VBS$, which is defined $\mu_{L, N}$-almost everywhere. Then $F \in L^2(\mu_{L, N})$, and \begin{equation*} \mu_{L, N}(F) = \frac{\langle \VBS, \varphi \rangle}{\| \VBS \|^2} , \qquad \mu_{L, N}(|F|^2) = \frac{\| \varphi \|^2}{\| \VBS \|^2} . \end{equation*} In particular, if $\varphi$ is orthogonal to $\VBS$, then \begin{equation}\label{e:groundvariance} \Var_{\mu_{L, N}}(F) = \frac{\| \varphi \|^2}{\| \VBS \|^2} . \end{equation} \end{lemma} \begin{proof} By the formula for $\mu_{L, N}$ in terms of $\iota \VBS$, we have \begin{equation*} \mu_{L, N}(|F|^2) = \frac{1}{\| \VBS \|^2} \int |\iota \varphi(s)|^2 \prod_{x \in \Lambda} 5 \dd s_x = \frac{\| \varphi \|^2}{\| \VBS \|^2} , \end{equation*} where we used that $\iota$ is an isometry. The formula for $\mu_{L, N}(F)$ follows in the same way from the identity $F |\iota \VBS|^2 = \overline{\iota \VBS} \, \iota \varphi$, which holds almost everywhere. \end{proof} \section{The block dynamics and the comparison theorem}\label{s:quantum-block} We now define the heat-bath kernels. Let $R \subset \Lambda$ and $s \in (\Sph)^\Lambda$. We let $\mathcal Q_R(s, \cdot)$ denote the probability measure on $(\Sph)^\Lambda$ under which the spins off $R$ are equal to $s_{\Lambda \setminus R}$, and the spins $\tilde s_R$ in $R$ have the density proportional to \begin{equation}\label{e:QR} \prod_{\{x, y\} \in \mathcal E(R)} (1 - \tilde s_x \cdot \tilde s_y) , \qquad \text{where } \tilde s_u = s_u \text{ for } u \in \Lambda \setminus R , \end{equation} with respect to $\prod_{x \in R} \dd \tilde s_x$. As a function of $\tilde s_R$, the product \eqref{e:QR} is continuous and nonnegative, with zeros only on a set of measure zero (where $\tilde s_x = \tilde s_y$ for some bond $\{x, y\} \in \mathcal E(R)$). In particular, its integral over $(\Sph)^R$ is positive for every $s$. We set \begin{equation*} (\mathcal Q_R F)(s) = \int F(\tilde s) \, \mathcal Q_R(s, \dd \tilde s) \end{equation*} for every bounded or nonnegative measurable function $F$ on $(\Sph)^\Lambda$. The density in \eqref{e:mu} is the product of the factors of the bonds in $\mathcal E(R)$, which form \eqref{e:QR}, and of factors that do not involve $s_R$. By this factorization and Fubini's theorem, $\mathcal Q_R F$ is a version of the conditional expectation of $F$ under $\mu_{L, N}$ given $s_{\Lambda \setminus R}$. In particular, $\mathcal Q_R$ extends to the orthogonal projection of $L^2(\mu_{L, N})$ onto the closed subspace of functions of $s_{\Lambda \setminus R}$. We call $\mathcal Q_R$ the \emph{heat-bath kernel} of $R$. For $F \in L^2(\mu_{L, N})$, we set $\Var_R F = \mathcal Q_R |F - \mathcal Q_R F|^2$, so that \begin{equation}\label{e:VarR} \mu_{L, N}(\Var_R F) = \| F - \mathcal Q_R F \|^2_{L^2(\mu_{L, N})} . \end{equation} Let $b \ge 1$ be an integer such that $L, N \ge b$. For $x \in \Lambda$, we call \begin{equation*} B_x = x + \{1, \dots, b\}^2 = \bigl\{ x + u : u \in \{1, \dots, b\}^2 \bigr\} \end{equation*} the \emph{block} with corner $x$; it has $b^2$ sites. We call the Markov dynamics on $(\Sph)^\Lambda$ in which, for every $x \in \Lambda$, the block $B_x$ is resampled from $\mathcal Q_{B_x}$ at rate $b^{-2}$ the \emph{block dynamics}. The generator of the block dynamics is \begin{equation}\label{e:blockgen} \Lbl = \frac{1}{b^2} \sum_{x \in \Lambda} (\mathcal Q_{B_x} - \Id) . \end{equation} The operator $\Lbl$ is bounded, self-adjoint, and nonpositive on $L^2(\mu_{L, N})$ and annihilates the constant functions. By \eqref{e:VarR}, its Dirichlet form is \begin{equation*} \langle F, -\Lbl F \rangle_{L^2(\mu_{L, N})} = \frac{1}{b^2} \sum_{x \in \Lambda} \mu_{L, N}(\Var_{B_x} F) , \qquad F \in L^2(\mu_{L, N}) . \end{equation*} We define its spectral gap $\gap(\Lbl)$ as in \zcref{s:intro-notation}. Next, we introduce the finite quantum system that enters the comparison. We call the set of sites $u \in \Z^2$ with $\dist(u, \{1, \dots, b\}^2) \le 1$ the \emph{patch}. It has $b^2 + 4b$ sites, and $2b^2 + 2b$ bonds of $\Z^2$ meet $\{1, \dots, b\}^2$. We let $h_b$ denote the sum of the projections $P_e$ over the bonds $e$ of $\Z^2$ that meet $\{1, \dots, b\}^2$, acting on the tensor product of copies of $V$ over the sites of the patch, where $P_e$ is defined as in \zcref{s:quantum-poly}. We call $h_b$ the \emph{patch operator}. Since $h_b$ is a nonzero nonnegative operator on a finite-dimensional space, it has a smallest positive eigenvalue, which we denote by $g_b$. Now let $L, N \ge b + 4$ and $x \in \Lambda$. The map $u \mapsto x + u$ is injective on the patch, since $L, N \ge b + 2$. It maps $\{1, \dots, b\}^2$ onto $B_x$ and the bonds of $\Z^2$ that meet $\{1, \dots, b\}^2$ bijectively onto $\mathcal E(B_x)$, since every bond of $\mathcal E(B_x)$ has the form $\{x + u, x + v\}$ with $u \in \{1, \dots, b\}^2$ and $v \sim u$. We identify the patch with its image under this map. Then $\mathcal H_\Lambda$ is the tensor product of the space of the patch and the space of the other sites of $\Lambda$, and $H_{B_x} = h_b \otimes \Id$. The operator $H_{B_x}$ commutes with $\Pi_{B_x}$, and the eigenvalues of $H_{B_x}$ on the range of $\Id - \Pi_{B_x}$ are positive eigenvalues of $h_b$. This implies that \begin{equation}\label{e:patchgap} H_{B_x} \ge g_b (\Id - \Pi_{B_x}) \qquad \text{for all } x \in \Lambda . \end{equation} An explicit lower bound for $g_b$ is proved in \zcref{l:patchgapexplicit}. We are now ready to state the main result of this section. \begin{theorem}\label{t:comparison} Let $b \ge 1$ be an integer, and let $L, N \ge b + 4$. Then \begin{equation*} \gap(H_{L, N}) \ge \frac{b}{b + 1} \, g_b \, \gap(\Lbl) . \end{equation*} \end{theorem} \begin{proof} Let $\varphi \in \mathcal H_\Lambda$ be nonzero and orthogonal to $\VBS$. We set $F = \iota \varphi / \iota \VBS$, and in this proof we write $\mu = \mu_{L, N}$ and $\| \cdot \|$ for the norm of $L^2(\mu)$. Recall that $\gap(H_{L, N})$ is the smallest nonzero eigenvalue of $H_{L, N}$. Using \zcref{c:kernel}, we see that it is the infimum of $\langle \varphi, H_{L, N} \varphi \rangle / \| \varphi \|^2_{\mathcal H_\Lambda}$ over all such $\varphi$. To prove the theorem, it suffices to show that \begin{equation}\label{e:comparison-goal} \gap(\Lbl) \, \| \varphi \|^2_{\mathcal H_\Lambda} \le \frac{b + 1}{b \, g_b} \langle \varphi, H_{L, N} \varphi \rangle . \end{equation} We begin with a single block, which is the main step. Fix $x \in \Lambda$, write $R = B_x$, and set $\psi_R = \Pi_R \varphi$. Then $H_R \psi_R = 0$, and $P_e \psi_R = 0$ for all $e \in \mathcal E(R)$, as in the proof of \zcref{c:kernel}. Applying \zcref{l:divisibility}, we find that $\psi_R = q_{\mathcal E(R)} \chi_R$ for a polynomial $\chi_R$ in the variables $(z_u)_{u \in \Lambda \setminus R}$. Observe that $\VBS = q_{\mathcal E(R)} \, q_{\mathcal E \setminus \mathcal E(R)}$, where $q_{\mathcal E \setminus \mathcal E(R)}$ is also a polynomial in these variables, since its bonds do not meet $R$. Dividing, we obtain \begin{equation}\label{e:ratio} \frac{\iota \psi_R(s)}{\iota \VBS(s)} = \frac{\chi_R(\zeta(s))}{q_{\mathcal E \setminus \mathcal E(R)}(\zeta(s))} \end{equation} for all $s$ with $\iota \VBS(s) \neq 0$, and hence for almost every $s$. The right side of \eqref{e:ratio} is a function of $s_{\Lambda \setminus R}$. It belongs to $L^2(\mu)$ by \zcref{l:coherent} applied to $\psi_R$. Recall that $\mathcal Q_R$ is the orthogonal projection onto the functions of $s_{\Lambda \setminus R}$ in $L^2(\mu)$. Using this fact together with \eqref{e:VarR}, \zcref{l:coherent} applied to $\varphi - \psi_R$, and \eqref{e:patchgap}, we obtain \begin{equation}\label{e:comparison-block} \begin{aligned} \mu(\Var_R F) &= \| F - \mathcal Q_R F \|^2 \le \Bigl\| F - \frac{\iota \psi_R}{\iota \VBS} \Bigr\|^2 = \frac{\| \varphi - \psi_R \|^2_{\mathcal H_\Lambda}}{\| \VBS \|^2_{\mathcal H_\Lambda}} \\ &= \frac{\langle \varphi, (\Id - \Pi_R) \varphi \rangle}{\| \VBS \|^2_{\mathcal H_\Lambda}} \le \frac{\langle \varphi, H_R \varphi \rangle}{g_b \, \| \VBS \|^2_{\mathcal H_\Lambda}} . \end{aligned} \end{equation} Next, we sum \eqref{e:comparison-block} over $x \in \Lambda$. We claim that every bond $e \in \mathcal E$ meets exactly $b^2 + b$ blocks. We may write $e = \{u, u + (1, 0)\}$ with $u \in \Lambda$ by symmetry. Using $L, N \ge b + 1$, we see that $u \in B_x$ for exactly $b^2$ sites $x$ and $u + (1, 0) \in B_x$ for exactly $b^2$ sites $x$. The two sites of $e$ lie in $B_x$ for exactly $b(b - 1)$ sites $x$, namely those for which $u = x + v$ with $v \in \{1, \dots, b - 1\} \times \{1, \dots, b\}$. By inclusion and exclusion, the number of blocks that meet $e$ is \begin{equation*} 2 b^2 - b(b - 1) = b^2 + b , \end{equation*} which proves the claim. Together with \eqref{e:HR}, the claim implies that \begin{equation*} \sum_{x \in \Lambda} H_{B_x} = (b^2 + b) H_{L, N} . \end{equation*} Summing \eqref{e:comparison-block} over $x$, dividing by $b^2$, and using this identity, we find that \begin{equation}\label{e:comparison-sum} \langle F, -\Lbl F \rangle_{L^2(\mu)} = \frac{1}{b^2} \sum_{x \in \Lambda} \mu(\Var_{B_x} F) \le \frac{b + 1}{b \, g_b} \, \frac{\langle \varphi, H_{L, N} \varphi \rangle}{\| \VBS \|^2_{\mathcal H_\Lambda}} . \end{equation} Finally, by \eqref{e:groundvariance}, the function $F$ is not constant, and \begin{equation*} \gap(\Lbl) \, \frac{\| \varphi \|^2_{\mathcal H_\Lambda}}{\| \VBS \|^2_{\mathcal H_\Lambda}} = \gap(\Lbl) \Var_\mu(F) \le \langle F, -\Lbl F \rangle_{L^2(\mu)} , \end{equation*} where the inequality follows from the definition of $\gap(\Lbl)$. Combining this bound with \eqref{e:comparison-sum}, we obtain \eqref{e:comparison-goal}, which completes the proof. \end{proof} \section{Local measures}\label{s:quantum-rotated} In the rest of the paper, we work with measures on configurations on finite subsets of $\Z^2$ in which the factors $1 - s_x \cdot s_y$ of \eqref{e:mu} are replaced by $1 + n_x \cdot n_y$. We now define these measures and relate them to the heat-bath kernels. Let $A \subset \Z^2$ be finite. We call a configuration $\eta \in (\Sph)^{\Z^2 \setminus A}$ a \emph{boundary configuration} of $A$, and we define the probability measure $\mu_A^\eta$ on $(\Sph)^A$ by \begin{equation}\label{e:muA} \mu_A^\eta(\dd n) = \frac{1}{Z_A^\eta} \prod_{\substack{\{x, y\} \subset A \\ x \sim y}} (1 + n_x \cdot n_y) \prod_{\substack{x \in A, \, y \notin A \\ x \sim y}} (1 + n_x \cdot \eta_y) \prod_{x \in A} \dd n_x , \end{equation} where $Z_A^\eta$ is the integral of the product of the factors over $(\Sph)^A$. The product of the factors vanishes only where $n_x = -n_y$ or $n_x = -\eta_y$ for one of the factors, which is a set of measure zero, so $Z_A^\eta > 0$ for every boundary configuration $\eta$. Note that $\mu_A^\eta$ is determined by the spins $\eta_y$ at the sites $y \notin A$ adjacent to $A$. For $W \subset A$, we let $\mu_{A, W}^\eta$ denote the marginal of $\mu_A^\eta$ on $(\Sph)^W$. We will need the following lemma in \zcref{s:block}. It shows that, after the spins on one of the two sublattices of $\Z^2$ are reversed, the heat-bath kernel of a block becomes a measure of the form \eqref{e:muA}. \begin{lemma}\label{l:rotated} Let $b \ge 1$ be an integer, let $L, N \ge b + 4$, let $x \in \Lambda$ and $c \in \Z^2$, and set $T = c + \{1, \dots, b\}^2 \subset \Z^2$. Let $s \in (\Sph)^\Lambda$, and let $\eta$ be a boundary configuration of $T$ such that \begin{equation*} \eta_u = (-1)^{u_1 + u_2} s_{x + u - c} \qquad \text{for all } u \in \Z^2 \setminus T \text{ adjacent to } T . \end{equation*} Let $\tilde s$ be a random configuration with law $\mathcal Q_{B_x}(s, \cdot)$. Then the random configuration $\bigl( (-1)^{u_1 + u_2} \tilde s_{x + u - c} \bigr)_{u \in T}$ has law $\mu_T^\eta$. \end{lemma} \begin{proof} The map $u \mapsto x + u - c$ is the map $u \mapsto x + u$ considered before \eqref{e:patchgap}, composed with the translation by $-c$. It follows that this map sends $T$ onto $B_x$ and the bonds of $\Z^2$ that meet $T$ bijectively onto $\mathcal E(B_x)$, and that it is injective on the set of sites at distance at most $1$ from $T$. We set \begin{equation*} n_u = (-1)^{u_1 + u_2} \tilde s_{x + u - c} , \qquad u \in T . \end{equation*} By \eqref{e:QR} and the invariance of the uniform measure on $\Sph$ under $a \mapsto -a$, the configuration $(n_u)_{u \in T}$ has a density with respect to $\prod_{u \in T} \dd n_u$ that is proportional to \begin{equation*} \prod_{\substack{\{u, v\} \text{ bond of } \Z^2 \\ \{u, v\} \cap T \neq \emptyset}} (1 - \tilde s_{x + u - c} \cdot \tilde s_{x + v - c}) . \end{equation*} For each $v \notin T$ adjacent to $T$, the site $x + v - c$ lies outside $B_x$ by injectivity on the sites at distance at most $1$ from $T$. Thus $\tilde s_{x + v - c} = s_{x + v - c}$ for each such $v$. Every bond $\{u, v\}$ of $\Z^2$ satisfies $(-1)^{u_1 + u_2} (-1)^{v_1 + v_2} = -1$, which gives \begin{equation*} 1 - \tilde s_{x + u - c} \cdot \tilde s_{x + v - c} = \begin{cases} 1 + n_u \cdot n_v & \text{if } u, v \in T , \\ 1 + n_u \cdot \eta_v & \text{if } u \in T \text{ and } v \notin T . \end{cases} \end{equation*} These are the factors in \eqref{e:muA} with $A = T$. \end{proof} The signs $(-1)^{u_1 + u_2}$ are defined on $\Z^2$ rather than on $\Lambda$, and each block is treated with its own translation. For this reason, \zcref{l:rotated} requires no parity condition on $L$ or $N$.